Advanced · Team and organisation · module 120 of 155
Project lesson: an implementation plan for your team
Bring data agreements, roles, test set, training and evaluation together in one implementation plan.
After this lesson
You write an implementation plan for one AI application in your team, with concrete agreements that someone else can follow and check.
In view · How it connects
An implementation plan someone else can follow
A traffic light for data, an owner and a point of contact.
Fixed test cases and training with examples from your own work.
Measuring points, an approach to incidents and a fixed review moment.
Part 1 of 3 · Data and roles
Your preference for less movement is on. Choose each part yourself.
What goes in the implementation plan
An implementation plan describes how one AI application is introduced in your team. It contains the purpose and the task, the approved tool, which data is and is not allowed (your traffic light from lesson 118), who does what and who decides. Add which legal points you had checked, such as the GDPR and the AI Act (lessons 116 and 117), and which questions are still open. Keep it short: a plan that fits on two pages actually gets read.
Roles, test set and training
Name an owner who decides on changes, a point of contact for questions and who follows up on errors (lesson 93). Add a small test set with ordinary cases and edge cases, so that after every change you can check whether the application still does what it should (lesson 113). Plan the training: what does every user need to know about the tool, the data agreements and the checking? This way you also take measures to support your team’s AI literacy.
Measure, evaluate and keep up to date
Record the baseline measurement and the metrics from your pilot, and the moment when you evaluate (lesson 119). Write down what happens in the event of an incident: who shuts down the application, who is notified and how do you continue working without AI? Agree when the plan will be reviewed. Have the plan read by a colleague who did not help write it: can they say without explanation what is allowed, who decides and what is measured? Then the plan is concrete enough.
New practice example
From explanation to a usable result.
A rollout plan brings together data agreements, roles, test set, training, measurement points and the incident procedure for one application.
Fictional practice situationFictional: a care home wants AI to help with the monthly newsletter for family members. A first plan only says: “We use AI for the newsletter and pay attention to privacy.”
View a possible solution
This is how it can be done
Completed plan: green for general activities and menus, red for residents’ names, photos and health information. Owner: the activities coordinator. Check: a second colleague proofreads every newsletter. Test set: three old newsletters as examples. Training: one hour with their own examples. Incident: the newsletter is only sent once it has been resolved. Review: after three issues.
Why this works
“Paying attention to privacy” was translated into concrete rules and roles. As a result, everyone can follow the plan and you can check it.
Make it slightly more difficult
Add one measurement point that shows whether the newsletter really becomes better or faster as a result.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
“Paying attention to privacy” was translated into concrete rules and roles. As a result, everyone can follow the plan and you can check it.
A step further
Add one measurement point that shows whether the newsletter really becomes better or faster as a result.
Try & discover · no obligation
Become the source checker yourself
Only mark statements that the source does not support. Then check your selection.
Fictional rollout plan for AI help with frequently asked questions in a library. Which parts can nobody check or carry out as they currently stand?
Select which sentences you would correct above. The answer remains hidden until you check your selection.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
Help me write a two-page implementation plan for this fictional application: library staff use an approved AI tool to prepare answers to visitors’ frequently asked questions. Use the headings purpose and task, tool, data (traffic light), roles, legal points to check, test set (six cases), training, metrics, incidents and review. Do not fill in unknown information, but mark it as an open point.
Apply it
Your implementation plan describes the tool and the training, but not who shuts down the application in the event of an incident or when the plan will be reviewed.
Which approach best fits this situation?
- Add an owner, an incident approach with a way of working without AI and a review moment.
- Leave it as it is; incidents rarely happen.
- Write that in case of problems everyone decides for themselves what needs to happen.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
Which part makes an implementation plan most verifiable?
- A general sentence saying that everyone works responsibly with AI.
- Concrete agreements on data, roles, a test set and metrics, with an evaluation moment.
- A list of all AI tools on the market.
Extra practice
Write the implementation plan for an application in your own team. Have a colleague who did not help write it read it, and note which questions they still had.
Purpose: faster and consistent answers to frequently asked questions. Data: green for public library information, red for borrowers’ data. Roles: the owner is the team leader, the point of contact is Jonas. Test set: six questions, including one about an unknown fine. Training: one hour with our own examples. Metrics: time per answer and number of corrections. Incident: shut down the tool, notify the team leader, continue manually. Review: after three months. Open point: have the legal basis confirmed.
